BERMAD: batch effect removal for single-cell RNA-seq data using a multi-layer adaptation autoencoder with...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional data crucial for biological discovery.
• Batch effects arising from different experimental platforms complicate data integration and analysis.
• Existing batch effect removal methods often struggle with under- or over-correction, particularly in nonlinear scRNA-seq data.

Purpose of the Study:

• To develop a robust method for accurate batch effect removal in scRNA-seq data integration.
• To address the limitations of existing methods, specifically under-correction and over-correction.
• To enhance the joint analysis of scRNA-seq datasets from diverse experimental origins.

Main Methods:

• Proposed a novel multi-layer adaptation autoencoder with a dual-channel framework, named BERMAD.
• Employed a multi-layer adaptation architecture to model batch distribution differences across feature granularities.
• Utilized a dual-channel framework with independently trained autoencoders to preserve batch-specific heterogeneous information.

Main Results:

• BERMAD demonstrates superior performance in scRNA-seq data integration compared to state-of-the-art methods.
• The method effectively mitigates both under-correction and over-correction issues in batch effect removal.
• Experiments confirm the ability of BERMAD to accurately correct batch effects while retaining biological signal heterogeneity.

Conclusions:

• BERMAD offers an effective solution for integrating and jointly analyzing scRNA-seq data from multiple sources.
• The proposed framework achieves more accurate batch correction by considering feature granularities and preserving biological variations.
• The method advances the field of scRNA-seq data integration, enabling more reliable downstream analyses.